[Ghaemmaghami AM, Hancock MJ, Harrington H, Kaji H, Khademhosseini A. Biomimetic tissues on a chip for drug discovery. Drug Discov Today. 2012 Feb;17(3-4):173-81. doi: 10.1016/j.drudis.2011.10.029. Epub 2011 Nov 7.]
Full Text: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3273552/
Developing biologically relevant models of human tissues and organs is an important enabling step for disease modeling and drug discovery. Recent advances in tissue engineering, biomaterials and microfluidics have led to the development of microscale functional units of such models also referred to as ‘organs on a chip’. In this review, we provide an overview of key enabling technologies and highlight the wealth of recent work regarding on-chip tissue models. In addition, we discuss the current challenges and future directions of organ-on-chip development.
A host of tissue models have been developed in academia and industry to mimic the sub-systems of organs or biological processes (Table 1 and Figure 2). In the following, we provide an overview of on-chip tissue models, their design, fabrication and physiological properties relevant to drug discovery. In particular, we highlight on-chip tissue models of the intestine, liver, lung and muscle, as well as models of tumors and blood vessels. We also outline recent attempts to combine the tissues of several organs on a single chip. Comprehensive reviews exist for bone, breast, cardiac, corneal, liver and tumor tissue models [51–53] as well as neural networks [54,55]; we expand on this literature by focusing more on liver and tumor tissue models. Other on-chip models that could be used for future studies on drug discovery and toxicity include those for sensory organs such as the nose [41,56,57] and biological processes such as sperm motility [42,43,58] and wound healing [59,60].
Intestine on a chip
Orally administered drugs are mainly absorbed in the small intestine, where they must diffuse across a mucous layer covering an epithelial cell layer lining the intestinal wall . Therefore, early in most development processes, drugs and chemicals are first screened against intestinal cells and tissues to assess their absorption, distribution, metabolism, elimination and toxicity (ADMET) . The small intestine epithelium is composed of two main cell types: enterocytes and goblet cells. Enterocytes constitute ~90% of the cell population and are covered with surface-enhancing microvilli. In most areas of the intestine, goblet cells constitute the remaining 10% of the cell population and secrete mucus which coats the epithelial layer and is important for innate immune responses and homeostasis. Coordinated intestinal movements (i.e. peristalsis) have also been shown to have an important role in intestinal biology. Every material absorbed through the intestine into the blood stream must first diffuse across the mucus layer, the enterocytes, the lamina propria and the capillaries formed by endothelial cells. Crossing the epithelial layer, however, has been shown to be the rate-limiting step .
Small intestine models destined for absorption and/or metabolism studies should include these key cellular (i.e. enterocytes, goblet and vascular endothelial cells), structural (i.e. villi and mucus) and dynamic (i.e. peristalsis) features. Some of these elements were addressed by Kimura et al. , who engineered an intestinal model with a membrane and vascular flow simulating the epithelial barrier and the epithelial-endothelial interface, respectively. Additional important model features include the mucus layer and the digestion process. These were partially addressed by Mahler et al., who developed a microscale cell culture analog of the gastrointestinal (GI) tract incorporating digestion, a mucus layer and physiologically realistic cell populations . To mimic drug transport across the intestinal wall better, additional structural features such as microscopic villi should be incorporated to enhance the intestinal epithelium surface area and to direct mucus. Such structural features have been fabricated in PDMS[2. AU: Please define] and hydrogels seeded with epithelial cells (Fig. 2j) , and could be incorporated into future functional tissue models.
Because healthy primary human cells are difficult to obtain, the majority of existing intestinal models were developed using immortalized cells lines. Although such cell lines can serve as informative and convenient surrogates for primary cells, their use limits the extrapolation of results to humans. Existing intestinal models also lack a host of ‘non-GI’ cells, including mast cells which play a key part in the barrier integrity and physiology of the intestinal epithelium .
Liver on a chip
Hepatotoxicity is a major side effect of many chemicals and drugs that are administered over prolonged periods of time. Almost half of all drug withdrawals occur as a result of acute liver toxicity . Therefore, efficient, reliable, accurate and inexpensive tools for liver toxicity testing are required. Recent reviews provide an overview of microfluidic devices incorporating liver tissue or cells for metabolism, drug discovery and toxicity studies [53,69]. Existing tools for testing liver toxicity, involving hepatocyte cell lines, two-dimensional (2D) in vitro cultures and animal models, all suffer from major limitations. Cell lines and 2D cultur
es lack key metabolic activities found in human liver and major interspecies differences limit the predictive value of animal models [70,71]. The key challenge for on-chip liver tissue models is to maintain the metabolic activity and phenotype of the poorly viable cryopreserved human primary hepatocytes. Bioreactors have been developed to meet this challenge, such as the perfused multiwell plate device of Domansky et al. , which supported the growth and functional integrity of hepatocytes in 3D culture for up to a week (Fig. 2h), and the multiwell micropatterned co-culture system of Khetani and Bhatia  which could maintain phenotypic functions for several weeks. Both groups co-cultured primary hepatocytes with other key liver cells such as sinusoidal endothelial cells containing stellate and Kupffer cells  or fibroblasts  to improve and maintain primary hepatocyte function considerably. Another desirable feature of liver tissue models is to simulate the morphology of the liver’s functional units (i.e. lobules) to provide the homogenous and heterogeneous cell–cell interactions and cellular matrix support to maintain hepatocyte functionality [72,73]. Toh et al.  incorporated this feature into their on-chip liver model that exhibited a reasonable simulation of the liver morphology and functionality for up to 72 hours. In a second example, Schütte et al. used a perfusion system and integrated electrodes to assemble liver sinusoids by dielectrophoresis (Fig. 2g) . A second class of on-chip liver models involves intact tissues extracted from humans or animals and integrated on-chip for the in vitro assessment of metabolism and toxicity . Recently, van Midwoudet al.  integrated liver and intestinal slices into different compartments of a microfluidic device with sequential perfusion between the compartments to study interorgan interactions. In addition to reflecting in vivo conditions adequately for investigating liver drug metabolism and toxicity, on-chip models incorporating primary hepatocytes or intact tissue slices must be able to maintain key metabolic activities and normal liver integrity, particularly with regard to repeat dose experiments . This feature was demonstrated by Khetani & Bhatia , who performed nine-day[3. AU: Is hyphen ok here? Added to draw a distinction between nine one-day repeat dose exps] repeat dose experiments with their co-culture system to test the chronic toxicity of troglitazone. Finally, an exciting new area is the ability to use in vitro cultures for disease modeling and to monitor disease progression and the impact of treatment in real-time. An example for such an approach is the recent work of Jones et al. , who used a fluorescent cell-based reporter system to monitor in real-time the infection of hepatitis C on hematoma cell lines and primary human hepatocytes.
Lung on a chip
The lung epithelium acts like a physical barrier between the host and the environment and responds to a variety of environmental insults such as pathogens and air pollution. The lung’s defense mechanisms include secreting antimicrobial peptides (e.g. defensins) and pro- and anti-inflammatory mediators, and recruiting innate and adaptive immune cells to the site of infection or damage. The efficiency of such protective responses as well as the homeostasis of the airways is underpinned by a variety of lung specialized epithelial cells (e.g. ciliated and mucus producing) and their crosstalk with other cell types such as antigen presenting cells, macrophages and myofibroblasts . The human lung also provides one of the largest vascularized interfaces between the host and the environment, and is therefore considered a convenient port of delivery for local and systemic drugs. Drug delivery studies tend to focus on drug permeability across the lung epithelial barrier. Within this context, most existing lung tissue models are limited to single cell culture representations of the alveolar epithelium comprising epithelial cell lines [78,79], limiting their physiological relevance to the human lung. Certain enhancements have been made to single cell culture models, such as microfluidics and flexible membranes to apply fluid and solid mechanical stresses to single cell alveolar models to simulate the effects of mechanical ventilation and physiologic or pathologic liquid plug flows [80–82].
More relevant lung tissue models destined for disease modeling and drug testing require additional cell types such as airway smooth muscle and immune cells, as well as key structural, functional and mechanical features of the human respiratory epithelium. Toward this goal, Huh et al. (Fig. 2b)  co-cultured alveolar epithelial, endothelial and certain immune cells on an artificial membrane and used adjacent vacuum channels to mimic the cyclic stretching during breathing. The inclusion of immune cells recapitulated certain aspects of human lung responses to infection and inflammation, enabling the assessment of the inflammatory response triggered by inhaled pathogens. Despite its many merits, this model used immortalized cell lines rather than human primary cells and did not include several other key structural or immune cells. It is not known whether the anatomical shape of the alveolar space or upper respiratory tract would have any bearing on the dynamics of cell–cell interaction. Future attempts, in addition to inclusion of primary cells, could focus on micropatterning scaffolds to improve the anatomical relevance and function of such tissue models.
Tumor on a chip
Developing anticancer therapies that target rapidly dividing cancer cells but leave normal healthy cells untouched is a grand challenge for cancer research. A recent review contains a comprehensive li
st of 3D tumor tissue models including multicellular spheroid, hollow fiber and multicellular layer models . Perfusion delivers potential therapeutic agents to such 3D tumor cell cultures, which emulates the heterogeneous blood supply delivered to tumors in vivo . In addition to these approaches, a popular strategy for developing anticancer drugs is HTS in which large combinatorial arrays of drug cocktails are produced and exposed to cells, which are then analyzed to assess the drug effects . Performing HTS within microfluidic devices enables smaller volumes of reagents to be used than in typical multiwell plate experiments. One key engineering challenge is to generate an array of drug concentrations on-chip. To meet this challenge, Jang et al. developed a microfluidic combinatorial system that produced 64 or 100 combinations of two chemical solutions at different concentrations and stored them in isolated chambers . The device inputs could be chosen such that the compositions of the stored mixtures spanned any triangle in the 2D space of chemical concentrations. A second engineering challenge is to expose different combinations or concentrations of drugs to cells cultured on-chip. Two groups recently accomplished this step by exposing human lung cancer cells to different concentrations of the anticancer drug verapamil (VP-16) [84,85]. In one study, the percentage of apoptotic cells in the verapamil-pretreated group was approximately double that of the control group, in agreement with previous flow cytometry analysis . Further HTS advances could be made by using 3D cultures which mimic in vivo conditions better than cell monolayers. For example, Fischbach et al.  cultured cancer cells in 3D alginate hydrogels and found that cell signaling was altered.
A second line of cancer research is to test anticancer therapies on-chip to optimize system parameters for different types of cancer cells, while requiring minute amounts of reagents and cells. For example, Jedrych et al.  created a microfluidics system for in vitro photodynamic therapy (PDT) . In PDT, light induces a photosensitizer delivered to the carcinoma cells, which in turn reacts with oxygen to produce a chemical toxic to the tumor cells. A key step in developing PDT is to optimize the system parameters such as the photosensitizer concentration for different types of cancer cells. Jedrych et al. demonstrated this step by testing different concentrations of photosensitizer on A549 cancer cells.
Vessels on a chip
Blood vessels are involved in most medical conditions and are thus integral components to organs-on-chip devices . A key engineering challenge is to grow blood vessels and capillaries on-chip. Dike et al. addressed this issue by using a surface patterned with 10 μm wide cell-adhesive lines surrounded by cell-repellent areas . Endothelial cells cultured on the cell-adhesive lines differentiated to form capillary tube-like structures containing central lumens. A different approach was carried out by Kobayashi et al. , who patterned endothelial cells along parallel hydrophilic lines and then transferred the cells to a hydrogel scaffold. Once on the scaffold, the cells changed their morphology to form capillary-like tubular structures. The capillary structures were tested by flowing fluorescent solution through them, and also by transplanting them into mice where blood cells were subsequently found in the lumen. To control the endothelial capillary tube formation (tubulogenesis) further, Raghavan et al. cultured endothelial cells within microscale channels filled with collagen hydrogel (Fig. 2i) . The cells were organized into tubes containing lumens with controlled branching and orientation. Future work must be done to generate capillary structures in vitro with similar mechanical properties to those in vivo.
A second major engineering challenge is to reproduce the microenvironment surrounding blood vessel walls, composed predominantly of vascular endothelial cells. A key component of this challenge is to understand how endothelial cells lining the vessel walls interact with other cells types. To address this issue, Chung et al.  and Kaji et al. [40,92] developed endothelium models to study the paracrine communication between endothelial and other cell types. Chung et al. found that MTLn3 cancer cells attracted endothelial cells to induce capillary formation, whereas smooth muscle cells suppressed endothelial activity . Kaji et al. used complimentary substrates to co-culture endothelial and HeLa (cervical cancer) cells and found, similarly, that paracrine factors secreted from one cell type and directed via the culture medium to the other cell type affected the migration behavior of both types [40,92]. Research has also been devoted toward understanding how the cyclic stretch due to vessel deformation regulates cell function, and how disruptions in these forces produce diseased states such as hypertension or arteriosclerosis . The effect of mechanical stimulation on endothelial cells in vitrohas been previously reviewed . An additional challenge is to apply chemical and mechanical stimuli simultaneously to cells to engineer more biologically relevant models. To address this issue, Song et al. developed a membrane-based microfluidic device that models the adhesion of breast cancer cells on an endothelium (formed on the porous membrane) during intravascular metastasis (Fig. 2d) . They showed that either the
addition of the chemokine CXCL12 from the basal side or the elevation of fluidic shear stress could enhance the metastatic cell adhesion. Many additional processes remain to be studied, including the chemical and cellular exchange across vessel walls, regulation of the permeability of surrounding tissues, coagulation of blood, transmigration of leukocytes and the regulation of vessel diameter.
Microfluidic platforms for probing the structure and function of actual blood vessels under physiological conditions also exist. Recently, Günther et al. developed a microfluidic platform enabling on-chip fixation, long-term culture, controlled delivery of drugs and automated acquisition of consecutive dose–response curves of intact mouse mesenteric artery segments (Fig. 2h) . The phenylephrine dose–response relationships produced by the platform were in good agreement with those obtained by conventional pressure myography.
Muscle on a chip
Skeletal muscle is one of the major insulin-target tissues responsible for the maintenance of whole body glucose homeostasis. Defects in the glucose uptake in skeletal muscle contribute to insulin resistance in type 2 diabetes. To investigate insulin- and contraction-induced glucose metabolism, on-chip skeletal tissue muscle models require in-vivo-like structure, electrode stimulation and co-culture with primary human skeletal cells with motor neurons. In-vivo-like structural features include myotube alignment and assembly into sarcomeres. On-chip myotube alignment has been guided by substrate pattern [97,98] and stiffness [99,100] and by electrical stimulation . An important aspect of a muscle tissue model is the ability to control muscle contraction. Such control has been accomplished by Tourovskaiaet al.  by stimulating myotubes with a laminar stream of agrin, a chemical secreted by neurons in vivo at the location where nerves contact the muscle. A different approach was developed by Kaji et al., who modulated the substrate stiffness with an atelocollagen and electrically stimulated the formed myotubes to demonstrate a positive correlation between the contractility of the myotubes and the glucose uptake. Control of individual myotubes was achieved by Nagamine et al. , who integrated a microelectrode array with aligned myotubes on a fibrin gel sheet (Fig. 2e). Future improvements to on-chip muscle tissue models include monitoring the glucose uptake and other metabolism events in real time and incorporating neuromuscular junctions by co-culturing myotubes with neurons and stimulating them to induce muscle contraction.
Multiple organs on a chip
Most existing on-chip tissue models represent a single organ, preventing investigations on a drug’s systemic effect. Microscale cell culture analog systems, also known as ‘animal-on-a-chip’, ‘human-on-a-chip’ or ‘body-on-a-chip’ systems, attempt to improve the prediction of the effects of drugs and toxicity on the whole body. In a typical multiorgan device, multiple cell types representing different organs are cultured in separate chambers on a single chip. The chambers are connected by channels based on their sequence in vivo to assess systematically the effects of drug action and metabolism in different organs. A number of multiorgan devices has been developed with cell lines representing various organs or tissues, including: liver, tumor and bone marrow based on a mathematical pharmacokinetics–pharmacodynamics model and used to test the toxicity of the anticancer drug 5-fluorouracil (Fig. 2f) ; intestine, liver and breast cancer cells to evaluate the intestinal absorption, hepatic metabolism and the antitarget cell bioactivity of drugs ; liver, lung, kidney and adipose tissue . Despite continued efforts and strong motivations to replace animal testing, these multiorgan systems are still in their infancy. The function of these systems is dictated by the cell types employed, and physiologically relevant cellular responses must be distinguished from artifacts due to in vitro cell culture. Also, barrier tissue analogs must be incorporated into existing multiorgan systems because these tissues can significantly reduce the bioavailability of drugs taken up through inhalation or absorbed through the skin. Finally, it is essential to benchmark the results of the on-chip systems with in vivo data under identical conditions and drug exposure. A more detailed discussion can be found in the literature by Eschet al. .